Beyond adaptive gradient: Fast-Controlled Minibatch Algorithm for large-scale optimization